bpp-core  2.1.0
bpp::MetaOptimizer Class Reference

Meta-optimizer. More...

#include <Bpp/Numeric/Function/MetaOptimizer.h>

+ Inheritance diagram for bpp::MetaOptimizer:
+ Collaboration diagram for bpp::MetaOptimizer:

List of all members.

Public Member Functions

 MetaOptimizer (Function *function, MetaOptimizerInfos *desc, unsigned int n=1)
 Build a new MetaOptimizer object.
virtual ~MetaOptimizer ()
 MetaOptimizer (const MetaOptimizer &opt)
MetaOptimizeroperator= (const MetaOptimizer &opt)
MetaOptimizerclone () const
 Create a copy of this object and send a pointer to it.
void setFunction (Function *function)
 Set the function to optimize.
void doInit (const ParameterList &parameters) throw (Exception)
 This function is called by the init() method and contains all calculations.
double doStep () throw (Exception)
 This function is called by the step() method and contains all calculations.
MetaOptimizerInfosgetOptimizers ()
const MetaOptimizerInfosgetOptimizers () const
void updateParameters (bool yn)
 Tell if we shall update all parameters after one optimization step.
bool updateParameters () const
 Tell if we shall update all parameters after one optimization step.
void setOptimizationProgressCharacter (const std::string &c)
 Set the character to be displayed during optimization.
const std::string & getOptimizationProgressCharacter () const
The Optimizer interface.
void init (const ParameterList &params) throw (Exception)
 Basic implementation.
double step () throw (Exception)
 Basic implementation.
double optimize () throw (Exception)
 Basic implementation.
bool isInitialized () const
const ParameterListgetParameters () const
double getParameterValue (const std::string &name) const
const FunctiongetFunction () const
 Get the current function being optimized.
FunctiongetFunction ()
 Get the current function being optimized.
bool hasFunction () const
 Tell if a funciton is associatied to this optimizer.
double getFunctionValue () const throw (NullPointerException)
 Get the current function value.
void setMessageHandler (OutputStream *mh)
 Set the message handler for this optimizer.
OutputStreamgetMessageHandler () const
void setProfiler (OutputStream *profiler)
 Set the profiler for this optimizer.
OutputStreamgetProfiler () const
unsigned int getNumberOfEvaluations () const
 Get the number of function evaluations performed since the call of the init function.
void setStopCondition (const OptimizationStopCondition &stopCondition)
 Set the stop condition of the optimization algorithm.
OptimizationStopConditiongetStopCondition ()
 Get the stop condition of the optimization algorithm.
const OptimizationStopConditiongetStopCondition () const
 Get the stop condition of the optimization algorithm.
OptimizationStopConditiongetDefaultStopCondition ()
 Get the default stop condition of the optimization algorithm.
const OptimizationStopConditiongetDefaultStopCondition () const
 Get the default stop condition of the optimization algorithm.
bool isToleranceReached () const
 Tell if the tolerance level is reached.
bool isMaximumNumberOfEvaluationsReached () const
 Tell if the maximum number of function evaluations is reached.
void setMaximumNumberOfEvaluations (unsigned int max)
 Set the maximum number of function evaluation to perform during optimization.
void setVerbose (unsigned int v)
 Set the verbose level.
unsigned int getVerbose () const
 Get the verbose level.
void setConstraintPolicy (const std::string &constraintPolicy)
 Set the constraint policy for this optimizer.
std::string getConstraintPolicy () const
 Get the constraint policy for this optimizer.
void addOptimizationListener (OptimizationListener *listener)
 Register a listener to this class.

Protected Member Functions

ParameterListgetParameters_ ()
ParametergetParameter_ (size_t i)
FunctiongetFunction_ ()
void setDefaultStopCondition_ (OptimizationStopCondition *osc)
Inner utilitary functions
void autoParameter ()
 Build a list of AutoParameter instead of Parameter.
void ignoreConstraints ()
 Remove the constraints of all the arguments.
void profile (double v)
 Print to the profile if there is one.
void profile (unsigned int v)
 Print to the profile if there is one.
void profile (const std::string &s)
 Print to the profile if there is one.
void profileln (double v)
 Print to the profile if there is one and end line.
void profileln (unsigned int v)
 Print to the profile if there is one and end line.
void profileln (const std::string &s)
 Print to the profile if there is one and end line.
void printPoint (const ParameterList &params, double value)
 Print parameters and corresponding function evaluation to profiler.
void printMessage (const std::string &message)
 Give a message to print to the message handler.
void fireOptimizationInitializationPerformed (const OptimizationEvent &event)
 Notify all listeners that optimizer initialization was performed.
void fireOptimizationStepPerformed (const OptimizationEvent &event)
 Notify all listeners that an optimization step was performed.
bool listenerModifiesParameters () const

Protected Attributes

unsigned int nbEvalMax_
 The maximum number of function evaluations allowed.
unsigned int nbEval_
 The current number of function evaluations achieved.
double currentValue_
 The current value of the function.
bool tolIsReached_
 Tell if the tolerance level has been reached.

Private Attributes

MetaOptimizerInfosoptDesc_
std::vector< ParameterListoptParameters_
std::vector< size_t > nbParameters_
unsigned int n_
double precisionStep_
unsigned int stepCount_
double initialValue_

Detailed Description

Meta-optimizer.

This optimizer uses a set of optimizers to applyied sequentially on distinct parameters. The set of optimizers is fully specified by a MetaOptimizerInfos object.

To decrease the optimization time, the precision of the optimizers can be increased progressively: if $\varepsilon$ is the final precision required, one may consider using a precision increment of $\sigma=\log_10(\varepsilon/n)$, where $n$ is the number of progressive steps. During the first step optimization step, the precisions of type 1 and 2 optimizers are set to $10^{\sigma}$, $10^{2\sigma}$ for step 2, ... until precision $10^{n\sigma}=\varepsilon$ at step $n$ and later. This saves some time spending in the first steps of the estimation. The number of steps $n$ is set in the constructor of the optimizer.

This optimizer can be used with numerical derivatives.

See also:
MetaOptimizerInfos.

Definition at line 191 of file MetaOptimizer.h.


Constructor & Destructor Documentation

MetaOptimizer::MetaOptimizer ( Function function,
MetaOptimizerInfos desc,
unsigned int  n = 1 
)

Build a new MetaOptimizer object.

Parameters:
functionThe function to be optimized.
descA MetaOptimizerInfos object describing the optimizers to use. The optimizer will own the instance of the MetaOptimizerInfos object.
nThe number of progressive steps to use in optimization).

Definition at line 58 of file MetaOptimizer.cpp.

References bpp::AbstractOptimizer::getDefaultStopCondition(), bpp::AbstractOptimizer::getStopCondition(), n_, precisionStep_, bpp::AbstractOptimizer::setDefaultStopCondition_(), bpp::AbstractOptimizer::setOptimizationProgressCharacter(), and bpp::AbstractOptimizer::setStopCondition().

Referenced by clone().

Definition at line 105 of file MetaOptimizer.cpp.

References optDesc_.

Definition at line 75 of file MetaOptimizer.cpp.


Member Function Documentation

void bpp::AbstractOptimizer::addOptimizationListener ( OptimizationListener listener) [inline, virtual, inherited]

Register a listener to this class.

All registered listeners will be informed when an optimization event occur. See the documentation of the class to know what kind of events are supported.

Parameters:
listenerA listener to be registered with.

Implements bpp::Optimizer.

Definition at line 227 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::listeners_.

MetaOptimizer* bpp::MetaOptimizer::clone ( ) const [inline, virtual]

Create a copy of this object and send a pointer to it.

Returns:
A pointer toward the copy object.

Implements bpp::Optimizer.

Definition at line 220 of file MetaOptimizer.h.

References MetaOptimizer().

void MetaOptimizer::doInit ( const ParameterList params) throw (Exception) [virtual]

This function is called by the init() method and contains all calculations.

Parameters:
paramsThe parameters to use for initialization.

Implements bpp::AbstractOptimizer.

Definition at line 113 of file MetaOptimizer.cpp.

References bpp::Optimizer::setConstraintPolicy(), bpp::Optimizer::setMessageHandler(), bpp::Optimizer::setProfiler(), and bpp::Optimizer::setVerbose().

void AbstractOptimizer::fireOptimizationInitializationPerformed ( const OptimizationEvent event) [protected, inherited]

Notify all listeners that optimizer initialization was performed.

This method should be called by the init method.

Parameters:
eventAn OptimizationEvent object.

Definition at line 313 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::listeners_.

void AbstractOptimizer::fireOptimizationStepPerformed ( const OptimizationEvent event) [protected, inherited]

Notify all listeners that an optimization step was performed.

This method should be called by the step method.

Parameters:
eventAn OptimizationEvent object.

Definition at line 323 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::listeners_.

Referenced by bpp::AbstractOptimizer::step().

std::string bpp::AbstractOptimizer::getConstraintPolicy ( ) const [inline, virtual, inherited]

Get the constraint policy for this optimizer.

Returns:
The constraint policy.

Implements bpp::Optimizer.

Definition at line 226 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::constraintPolicy_.

const OptimizationStopCondition* bpp::AbstractOptimizer::getDefaultStopCondition ( ) const [inline, virtual, inherited]

Get the default stop condition of the optimization algorithm.

Returns:
The default stop condition used while optimizing.

Implements bpp::Optimizer.

Definition at line 219 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::defaultStopCondition_.

Function* bpp::AbstractOptimizer::getFunction ( ) [inline, virtual, inherited]

Get the current function being optimized.

Returns:
A const pointer toward the function being optimized.

Implements bpp::Optimizer.

Reimplemented in bpp::BfgsMultiDimensions, bpp::ConjugateGradientMultiDimensions, and bpp::NewtonOneDimension.

Definition at line 198 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::function_.

double bpp::AbstractOptimizer::getFunctionValue ( ) const throw (NullPointerException) [inline, virtual, inherited]

Get the current function value.

Returns:
The value of the function at the point specified by _parameters.
Exceptions:
NullPointerExceptionIf no function is associated with this optimizer.

Implements bpp::Optimizer.

Reimplemented in bpp::GoldenSectionSearch.

Definition at line 200 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::currentValue_, and bpp::AbstractOptimizer::function_.

OutputStream* bpp::AbstractOptimizer::getMessageHandler ( ) const [inline, virtual, inherited]
Returns:
The stream used for handling messages, if any.

Implements bpp::Optimizer.

Definition at line 207 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::messageHandler_.

Referenced by bpp::PowellMultiDimensions::doStep().

unsigned int bpp::AbstractOptimizer::getNumberOfEvaluations ( ) const [inline, virtual, inherited]

Get the number of function evaluations performed since the call of the init function.

Returns:
The number of function evaluations.

Implements bpp::Optimizer.

Definition at line 211 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::nbEval_.

Referenced by bpp::SimpleNewtonMultiDimensions::doStep(), bpp::SimpleMultiDimensions::doStep(), bpp::OneDimensionOptimizationTools::lineMinimization(), and bpp::OneDimensionOptimizationTools::lineSearch().

const std::string& bpp::AbstractOptimizer::getOptimizationProgressCharacter ( ) const [inline, inherited]
Returns:
The character to be displayed during optimization.

Definition at line 265 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::stepChar_.

Returns:
The MetaOptimizerInfos object associated to this optimizer.

Definition at line 238 of file MetaOptimizer.h.

References optDesc_.

Returns:
The MetaOptimizerInfos object associated to this optimizer.

Definition at line 243 of file MetaOptimizer.h.

References optDesc_.

double bpp::AbstractOptimizer::getParameterValue ( const std::string &  name) const [inline, virtual, inherited]
Returns:
The value associated with a given parameter name.

Implements bpp::Optimizer.

Definition at line 191 of file AbstractOptimizer.h.

References bpp::ParameterList::getParameterValue(), and bpp::AbstractOptimizer::parameters_.

OutputStream* bpp::AbstractOptimizer::getProfiler ( ) const [inline, virtual, inherited]
Returns:
The stream used for profiling, if any.

Implements bpp::Optimizer.

Definition at line 209 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::profiler_.

const OptimizationStopCondition* bpp::AbstractOptimizer::getStopCondition ( ) const [inline, virtual, inherited]

Get the stop condition of the optimization algorithm.

Returns:
The stop condition used while optimizing.

Implements bpp::Optimizer.

Definition at line 217 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::stopCondition_.

bool bpp::AbstractOptimizer::hasFunction ( ) const [inline, virtual, inherited]

Tell if a funciton is associatied to this optimizer.

Returns:
True if a function has been associated to this optimizer.

Implements bpp::Optimizer.

Definition at line 199 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::function_.

Referenced by bpp::GoldenSectionSearch::getFunctionValue().

void AbstractOptimizer::ignoreConstraints ( ) [protected, inherited]

Remove the constraints of all the arguments.

Definition at line 303 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::parameters_, and bpp::ParameterList::size().

Referenced by bpp::AbstractOptimizer::AbstractOptimizer(), and bpp::AbstractOptimizer::operator=().

void AbstractOptimizer::init ( const ParameterList params) throw (Exception) [virtual, inherited]

Basic implementation.

Store all parameters, call the doInit method, print to profiler, initialize timer and notify all listeners.

Implements bpp::Optimizer.

Definition at line 153 of file AbstractOptimizer.cpp.

References bpp::AutoParameter::CONSTRAINTS_AUTO, and bpp::AutoParameter::CONSTRAINTS_IGNORE.

Referenced by bpp::SimpleNewtonMultiDimensions::doStep(), bpp::SimpleMultiDimensions::doStep(), bpp::OneDimensionOptimizationTools::lineMinimization(), and bpp::OneDimensionOptimizationTools::lineSearch().

bool bpp::AbstractOptimizer::isInitialized ( ) const [inline, virtual, inherited]
Returns:
'true' if this optimizer has been initialized.

Implements bpp::Optimizer.

Definition at line 189 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::isInitialized_.

bool bpp::AbstractOptimizer::isMaximumNumberOfEvaluationsReached ( ) const [inline, virtual, inherited]

Tell if the maximum number of function evaluations is reached.

Returns:
Whether the maximum number of function evaluations is reached or not.

Implements bpp::Optimizer.

Definition at line 221 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::nbEval_, and bpp::AbstractOptimizer::nbEvalMax_.

bool bpp::AbstractOptimizer::isToleranceReached ( ) const [inline, virtual, inherited]

Tell if the tolerance level is reached.

Returns:
Whether the tolerance is reached or not.
See also:
OptimizationStopCondition

Implements bpp::Optimizer.

Definition at line 220 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::tolIsReached_.

bool AbstractOptimizer::listenerModifiesParameters ( ) const [protected, inherited]

Definition at line 333 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::listeners_.

Referenced by bpp::AbstractOptimizer::step().

MetaOptimizer & MetaOptimizer::operator= ( const MetaOptimizer opt)
void AbstractOptimizer::printMessage ( const std::string &  message) [protected, inherited]

Give a message to print to the message handler.

Parameters:
messageThe message to print.

Definition at line 284 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::messageHandler_.

Referenced by bpp::NewtonOneDimension::doStep(), and bpp::BfgsMultiDimensions::doStep().

void AbstractOptimizer::printPoint ( const ParameterList params,
double  value 
) [protected, inherited]

Print parameters and corresponding function evaluation to profiler.

Parameters:
paramsThe parameters to print.
valueThe function evaluation.

Definition at line 265 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::nbEval_, bpp::AbstractOptimizer::profile(), bpp::AbstractOptimizer::profileln(), bpp::ParameterList::size(), bpp::AbstractOptimizer::startTime_, and bpp::TextTools::toString().

Referenced by bpp::PowellMultiDimensions::doStep(), and bpp::AbstractOptimizer::step().

void AbstractOptimizer::profile ( double  v) [protected, inherited]

Print to the profile if there is one.

Parameters:
vThe double value to print.

Definition at line 224 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::profiler_.

Referenced by bpp::AbstractOptimizer::printPoint().

void AbstractOptimizer::profile ( unsigned int  v) [protected, inherited]

Print to the profile if there is one.

Parameters:
vThe unsigned int value to print.

Definition at line 238 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::profiler_.

void AbstractOptimizer::profile ( const std::string &  s) [protected, inherited]

Print to the profile if there is one.

Parameters:
sThe string to print to the profile.

Definition at line 251 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::profiler_.

void AbstractOptimizer::profileln ( double  v) [protected, inherited]

Print to the profile if there is one and end line.

Parameters:
vThe double value to print.

Definition at line 231 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::profiler_.

Referenced by bpp::AbstractOptimizer::printPoint().

void AbstractOptimizer::profileln ( unsigned int  v) [protected, inherited]

Print to the profile if there is one and end line.

Parameters:
vThe unsigned int value to print.

Definition at line 244 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::profiler_.

void AbstractOptimizer::profileln ( const std::string &  s) [protected, inherited]

Print to the profile if there is one and end line.

Parameters:
sThe string to print to the profile.

Definition at line 258 of file AbstractOptimizer.cpp.

References bpp::AbstractOptimizer::profiler_.

void bpp::AbstractOptimizer::setConstraintPolicy ( const std::string &  constraintPolicy) [inline, virtual, inherited]

Set the constraint policy for this optimizer.

Parameters:
constraintPolicyThe constraint policy.

Implements bpp::Optimizer.

Definition at line 225 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::constraintPolicy_.

Referenced by bpp::OneDimensionOptimizationTools::lineMinimization(), and bpp::OneDimensionOptimizationTools::lineSearch().

void bpp::MetaOptimizer::setFunction ( Function function) [inline, virtual]

Set the function to optimize.

Parameters:
functionThe function to optimize.

Reimplemented from bpp::AbstractOptimizer.

Definition at line 224 of file MetaOptimizer.h.

References bpp::MetaOptimizerInfos::getNumberOfOptimizers(), bpp::MetaOptimizerInfos::getOptimizer(), optDesc_, and bpp::Optimizer::setFunction().

void bpp::AbstractOptimizer::setMaximumNumberOfEvaluations ( unsigned int  max) [inline, virtual, inherited]

Set the maximum number of function evaluation to perform during optimization.

Parameters:
maxThe maximum number of evaluations to perform.

Implements bpp::Optimizer.

Definition at line 222 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::nbEvalMax_.

Referenced by bpp::BrentOneDimension::BrentOneDimension(), and bpp::NewtonBacktrackOneDimension::NewtonBacktrackOneDimension().

void bpp::AbstractOptimizer::setMessageHandler ( OutputStream mh) [inline, virtual, inherited]

Set the message handler for this optimizer.

The message handler keeps all messages that the optimizer may send. The default handler is set to standard output, but you can pass any ostream object (cerr, ofstream, etc.).

A NULL pointer disables message output.

Parameters:
mhThe message handler to use.

Implements bpp::Optimizer.

Definition at line 206 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::messageHandler_.

Referenced by bpp::OneDimensionOptimizationTools::lineMinimization(), and bpp::OneDimensionOptimizationTools::lineSearch().

void bpp::AbstractOptimizer::setProfiler ( OutputStream profiler) [inline, virtual, inherited]

Set the profiler for this optimizer.

The profiler keeps all the intermediate values taken by the parameters. The default profiler is set to standard output, but you can pass any ostream object (cerr, ofstream, etc.).

A NULL pointer disables message output.

Parameters:
profilerThe profiler to use.

Implements bpp::Optimizer.

Definition at line 208 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::profiler_.

Referenced by bpp::OneDimensionOptimizationTools::lineMinimization(), and bpp::OneDimensionOptimizationTools::lineSearch().

void bpp::AbstractOptimizer::setVerbose ( unsigned int  v) [inline, virtual, inherited]

Set the verbose level.

0 = off 1 = on 2 = more verbose 3 = even more, etc.

In most cases, only the 0 and 1 levels are implemented.

Parameters:
vverbose level.

Implements bpp::Optimizer.

Definition at line 223 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::verbose_.

Referenced by bpp::OneDimensionOptimizationTools::lineMinimization(), and bpp::OneDimensionOptimizationTools::lineSearch().

void bpp::AbstractOptimizer::updateParameters ( bool  yn) [inline, inherited]

Tell if we shall update all parameters after one optimization step.

This is required only for functions that have non-independent parameters, which means that setting one parameter value may modify one or several other parameters. Depending on the optimizer, this may have no effect.

Parameters:
yntrue/false

Definition at line 243 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::updateParameters_.

bool bpp::AbstractOptimizer::updateParameters ( ) const [inline, inherited]

Tell if we shall update all parameters after one optimization step.

This is required only for functions that have non-independent parameters, which means that setting one parameter value may modify one or several other parameters. Depending on the optimizer, this may have no effect.

Returns:
yn true/false

Definition at line 254 of file AbstractOptimizer.h.

References bpp::AbstractOptimizer::updateParameters_.


Member Data Documentation

Definition at line 201 of file MetaOptimizer.h.

Referenced by doStep(), and operator=().

unsigned int bpp::MetaOptimizer::n_ [private]

Definition at line 198 of file MetaOptimizer.h.

Referenced by doStep(), MetaOptimizer(), and operator=().

std::vector<size_t> bpp::MetaOptimizer::nbParameters_ [private]

Definition at line 197 of file MetaOptimizer.h.

Referenced by doStep(), and operator=().

Definition at line 196 of file MetaOptimizer.h.

Referenced by doStep(), and operator=().

Definition at line 199 of file MetaOptimizer.h.

Referenced by doStep(), MetaOptimizer(), and operator=().

unsigned int bpp::MetaOptimizer::stepCount_ [private]

Definition at line 200 of file MetaOptimizer.h.

Referenced by doStep(), and operator=().


The documentation for this class was generated from the following files:
 All Classes Namespaces Files Functions Variables Typedefs Friends